It's a good demonstration that partially-fmv games could leverage this to blend fmv and rendered scenes more if you trained it on your actors and sets. You would need significantly higher fidelity facial animations, however.

For anything even slightly less realistic, ew no.